Realtime_Multi-Person_Pose_Estimation

Why we included this project

This is the original research code behind Part Affinity Fields, the method that won the 2016 MSCOCO Keypoints Challenge and later grew into OpenPose. Because it works bottom-up, it skips person detection entirely and finds every person's joints in a single pass, which makes it a clear way to study how realtime multi-person pose estimation is built. The repo carries Caffe training and testing code along with demo paths in MATLAB, Python, and Jupyter, so a student can follow the full pipeline from COCO data preparation to network training. Teams that want a deployable production library are better off with the C++ successor it links to, but for understanding the architecture, this original version beats the reimplementations.